2 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
4 * This file is part of Libav.
6 * Libav is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
11 * Libav is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with Libav; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21 #ifndef AVCODEC_X86_CABAC_H
22 #define AVCODEC_X86_CABAC_H
24 #include "libavcodec/cabac.h"
25 #include "libavutil/attributes.h"
26 #include "libavutil/x86_cpu.h"
30 #define BRANCHLESS_GET_CABAC_UPDATE(ret, cabac, statep, low, lowword, range, tmp)\
31 "mov "tmp" , %%ecx \n\t"\
32 "shl $17 , "tmp" \n\t"\
33 "cmp "low" , "tmp" \n\t"\
34 "cmova %%ecx , "range" \n\t"\
35 "sbb %%ecx , %%ecx \n\t"\
36 "and %%ecx , "tmp" \n\t"\
37 "xor %%ecx , "ret" \n\t"\
38 "sub "tmp" , "low" \n\t"
39 #else /* HAVE_FAST_CMOV */
40 #define BRANCHLESS_GET_CABAC_UPDATE(ret, cabac, statep, low, lowword, range, tmp)\
41 "mov "tmp" , %%ecx \n\t"\
42 "shl $17 , "tmp" \n\t"\
43 "sub "low" , "tmp" \n\t"\
44 "sar $31 , "tmp" \n\t" /*lps_mask*/\
45 "sub %%ecx , "range" \n\t" /*RangeLPS - range*/\
46 "and "tmp" , "range" \n\t" /*(RangeLPS - range)&lps_mask*/\
47 "add %%ecx , "range" \n\t" /*new range*/\
48 "shl $17 , %%ecx \n\t"\
49 "and "tmp" , %%ecx \n\t"\
50 "sub %%ecx , "low" \n\t"\
51 "xor "tmp" , "ret" \n\t"
52 #endif /* HAVE_FAST_CMOV */
54 #define BRANCHLESS_GET_CABAC(ret, cabac, statep, low, lowword, range, tmp, tmpbyte, byte) \
55 "movzbl "statep" , "ret" \n\t"\
56 "mov "range" , "tmp" \n\t"\
57 "and $0xC0 , "range" \n\t"\
58 "movzbl "MANGLE(ff_h264_lps_range)"("ret", "range", 2), "range" \n\t"\
59 "sub "range" , "tmp" \n\t"\
60 BRANCHLESS_GET_CABAC_UPDATE(ret, cabac, statep, low, lowword, \
62 "movzbl " MANGLE(ff_h264_norm_shift) "("range"), %%ecx \n\t"\
63 "shl %%cl , "range" \n\t"\
64 "movzbl "MANGLE(ff_h264_mlps_state)"+128("ret"), "tmp" \n\t"\
65 "shl %%cl , "low" \n\t"\
66 "mov "tmpbyte" , "statep" \n\t"\
67 "test "lowword" , "lowword" \n\t"\
69 "mov "byte"("cabac"), %%"REG_c" \n\t"\
70 "add"OPSIZE" $2 , "byte "("cabac") \n\t"\
71 "movzwl (%%"REG_c") , "tmp" \n\t"\
72 "lea -1("low") , %%ecx \n\t"\
73 "xor "low" , %%ecx \n\t"\
74 "shr $15 , %%ecx \n\t"\
76 "shr $15 , "tmp" \n\t"\
77 "movzbl " MANGLE(ff_h264_norm_shift) "(%%ecx), %%ecx \n\t"\
78 "sub $0xFFFF , "tmp" \n\t"\
80 "add $7 , %%ecx \n\t"\
81 "shl %%cl , "tmp" \n\t"\
82 "add "tmp" , "low" \n\t"\
85 #if ARCH_X86 && HAVE_7REGS && !defined(BROKEN_RELOCATIONS)
86 #define get_cabac_inline get_cabac_inline_x86
87 static av_always_inline int get_cabac_inline_x86(CABACContext *c,
90 int bit, low, range, tmp;
93 "movl %a6(%5), %2 \n\t"
94 "movl %a7(%5), %1 \n\t"
95 BRANCHLESS_GET_CABAC("%0", "%5", "(%4)", "%1", "%w1", "%2",
97 "movl %2, %a6(%5) \n\t"
98 "movl %1, %a7(%5) \n\t"
100 :"=&r"(bit), "=&r"(low), "=&r"(range), "=&q"(tmp)
102 "i"(offsetof(CABACContext, range)), "i"(offsetof(CABACContext, low)),
103 "i"(offsetof(CABACContext, bytestream))
108 #endif /* ARCH_X86 && HAVE_7REGS && !defined(BROKEN_RELOCATIONS) */
110 #define get_cabac_bypass_sign get_cabac_bypass_sign_x86
111 static av_always_inline int get_cabac_bypass_sign_x86(CABACContext *c, int val)
115 "movl %a3(%2), %k1 \n\t"
116 "movl %a4(%2), %%eax \n\t"
118 "add %%eax, %%eax \n\t"
119 "sub %k1, %%eax \n\t"
121 "and %%edx, %k1 \n\t"
122 "add %k1, %%eax \n\t"
123 "xor %%edx, %%ecx \n\t"
124 "sub %%edx, %%ecx \n\t"
125 "test %%ax, %%ax \n\t"
127 "mov %a5(%2), %1 \n\t"
128 "subl $0xFFFF, %%eax \n\t"
129 "movzwl (%1), %%edx \n\t"
131 "shrl $15, %%edx \n\t"
133 "addl %%edx, %%eax \n\t"
134 "mov %1, %a5(%2) \n\t"
136 "movl %%eax, %a4(%2) \n\t"
138 :"+c"(val), "=&r"(tmp)
140 "i"(offsetof(CABACContext, range)), "i"(offsetof(CABACContext, low)),
141 "i"(offsetof(CABACContext, bytestream))
142 : "%eax", "%edx", "memory"
147 #endif /* AVCODEC_X86_CABAC_H */